From 9570dddac3a7cf74e86a45f2cfc35e2fedc1d926 Mon Sep 17 00:00:00 2001 From: Johannes Schriewer Date: Wed, 28 Feb 2024 21:25:27 +0100 Subject: [PATCH] SpeexDSP resampler --- audio/deps/resampler/AUTHORS | 18 + audio/deps/resampler/COPYING | 35 + audio/deps/resampler/README.md | 6 + audio/deps/resampler/arch.h | 101 +++ audio/deps/resampler/config.h | 18 + audio/deps/resampler/fixed_arm4.h | 135 +++ audio/deps/resampler/fixed_arm5e.h | 160 ++++ audio/deps/resampler/fixed_debug.h | 497 +++++++++++ audio/deps/resampler/fixed_generic.h | 106 +++ audio/deps/resampler/os_support.h | 89 ++ audio/deps/resampler/resample.c | 1051 ++++++++++++++++++++++++ audio/deps/resampler/resample_neon.h | 188 +++++ audio/deps/resampler/speex_resampler.h | 216 +++++ audio/deps/resampler/speexdsp_types.h | 28 + 14 files changed, 2648 insertions(+) create mode 100644 audio/deps/resampler/AUTHORS create mode 100644 audio/deps/resampler/COPYING create mode 100644 audio/deps/resampler/README.md create mode 100644 audio/deps/resampler/arch.h create mode 100644 audio/deps/resampler/config.h create mode 100644 audio/deps/resampler/fixed_arm4.h create mode 100644 audio/deps/resampler/fixed_arm5e.h create mode 100644 audio/deps/resampler/fixed_debug.h create mode 100644 audio/deps/resampler/fixed_generic.h create mode 100644 audio/deps/resampler/os_support.h create mode 100644 audio/deps/resampler/resample.c create mode 100644 audio/deps/resampler/resample_neon.h create mode 100644 audio/deps/resampler/speex_resampler.h create mode 100644 audio/deps/resampler/speexdsp_types.h diff --git a/audio/deps/resampler/AUTHORS b/audio/deps/resampler/AUTHORS new file mode 100644 index 0000000..395c3fe --- /dev/null +++ b/audio/deps/resampler/AUTHORS @@ -0,0 +1,18 @@ +Jean-Marc Valin + All the code except the following + +David Rowe + lsp.c lsp.h + Also ideas and feedback + +John Francis Edwards + wave_out.[ch], some #ifdefs for windows port and MSVC project files + +Segher Boessenkool + Misc. optimizations (for QMF in particular) + +Atsuhiko Yamanaka : + Patch to speexenc.c to add Vorbis comment format + +Radim Kolar : + Patch to speexenc.c for supporting more input formats diff --git a/audio/deps/resampler/COPYING b/audio/deps/resampler/COPYING new file mode 100644 index 0000000..7d6dc4a --- /dev/null +++ b/audio/deps/resampler/COPYING @@ -0,0 +1,35 @@ +Copyright 2002-2008 Xiph.org Foundation +Copyright 2002-2008 Jean-Marc Valin +Copyright 2005-2007 Analog Devices Inc. +Copyright 2005-2008 Commonwealth Scientific and Industrial Research + Organisation (CSIRO) +Copyright 1993, 2002, 2006 David Rowe +Copyright 2003 EpicGames +Copyright 1992-1994 Jutta Degener, Carsten Bormann + +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: + +- Redistributions of source code must retain the above copyright +notice, this list of conditions and the following disclaimer. + +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. + +- Neither the name of the Xiph.org Foundation nor the names of its +contributors may be used to endorse or promote products derived from +this software without specific prior written permission. + +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR +CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, +EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, +PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR +PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF +LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING +NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS +SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. diff --git a/audio/deps/resampler/README.md b/audio/deps/resampler/README.md new file mode 100644 index 0000000..367862c --- /dev/null +++ b/audio/deps/resampler/README.md @@ -0,0 +1,6 @@ +# SpeexDSP Resampler + +This code was lifted from `speexdsp-1.2.1` and modified to be as small as possible. +Only the fixed point API remains, all external dependencies to other files from +speexdsp have been included so this just contains the resampler. + diff --git a/audio/deps/resampler/arch.h b/audio/deps/resampler/arch.h new file mode 100644 index 0000000..353857d --- /dev/null +++ b/audio/deps/resampler/arch.h @@ -0,0 +1,101 @@ +/* Copyright (C) 2003 Jean-Marc Valin */ +/** + @file arch.h + @brief Various architecture definitions Speex +*/ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + - Neither the name of the Xiph.org Foundation nor the names of its + contributors may be used to endorse or promote products derived from + this software without specific prior written permission. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR + CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef ARCH_H +#define ARCH_H + +/* A couple test to catch stupid option combinations */ +#if defined(ARM4_ASM) + defined(ARM5E_ASM) + defined(BFIN_ASM) > 1 +#error Make up your mind. What CPU do you have? +#endif + +#include "speexdsp_types.h" + +#define ABS(x) ((x) < 0 ? (-(x)) : (x)) /**< Absolute integer value. */ +#define ABS16(x) ((x) < 0 ? (-(x)) : (x)) /**< Absolute 16-bit value. */ +#define MIN16(a,b) ((a) < (b) ? (a) : (b)) /**< Maximum 16-bit value. */ +#define MAX16(a,b) ((a) > (b) ? (a) : (b)) /**< Maximum 16-bit value. */ +#define ABS32(x) ((x) < 0 ? (-(x)) : (x)) /**< Absolute 32-bit value. */ +#define MIN32(a,b) ((a) < (b) ? (a) : (b)) /**< Maximum 32-bit value. */ +#define MAX32(a,b) ((a) > (b) ? (a) : (b)) /**< Maximum 32-bit value. */ + +typedef spx_int16_t spx_word16_t; +typedef spx_int32_t spx_word32_t; +typedef spx_word32_t spx_mem_t; +typedef spx_word16_t spx_coef_t; +typedef spx_word16_t spx_lsp_t; +typedef spx_word32_t spx_sig_t; + +#define Q15ONE 32767 + +#define LPC_SCALING 8192 +#define SIG_SCALING 16384 +#define LSP_SCALING 8192. +#define GAMMA_SCALING 32768. +#define GAIN_SCALING 64 +#define GAIN_SCALING_1 0.015625 + +#define LPC_SHIFT 13 +#define LSP_SHIFT 13 +#define SIG_SHIFT 14 +#define GAIN_SHIFT 6 + +#define WORD2INT(x) ((x) < -32767 ? -32768 : ((x) > 32766 ? 32767 : (x))) + +#define VERY_SMALL 0 +#define VERY_LARGE32 ((spx_word32_t)2147483647) +#define VERY_LARGE16 ((spx_word16_t)32767) +#define Q15_ONE ((spx_word16_t)32767) + + +#ifdef FIXED_DEBUG +# include "fixed_debug.h" +#else +# include "fixed_generic.h" + +# ifdef ARM5E_ASM +# include "fixed_arm5e.h" +# elif defined(ARM4_ASM) +# include "fixed_arm4.h" +# endif + +#endif + + +#ifdef FIXED_DEBUG +extern long long spx_mips; +#endif + + +#endif /* ARCH_H */ diff --git a/audio/deps/resampler/config.h b/audio/deps/resampler/config.h new file mode 100644 index 0000000..61cb173 --- /dev/null +++ b/audio/deps/resampler/config.h @@ -0,0 +1,18 @@ +/* Make use of ARM4 assembly optimizations */ +/* #undef ARM4_ASM */ + +/* Make use of ARM5E assembly optimizations */ +/* #undef ARM5E_ASM */ + +/* Enable NEON support */ +#define USE_NEON /**/ + +/* Symbol visibility prefix */ +#define EXPORT __attribute__((visibility("default"))) + +/* Debug fixed-point implementation */ +/* #undef FIXED_DEBUG */ + +/* Use C99 variable-size arrays */ +#define VAR_ARRAYS /**/ + diff --git a/audio/deps/resampler/fixed_arm4.h b/audio/deps/resampler/fixed_arm4.h new file mode 100644 index 0000000..a80fadb --- /dev/null +++ b/audio/deps/resampler/fixed_arm4.h @@ -0,0 +1,135 @@ +/* Copyright (C) 2004 Jean-Marc Valin */ +/** + @file fixed_arm4.h + @brief ARM4 fixed-point operations +*/ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + - Neither the name of the Xiph.org Foundation nor the names of its + contributors may be used to endorse or promote products derived from + this software without specific prior written permission. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR + CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef FIXED_ARM4_H +#define FIXED_ARM4_H + +#undef MULT16_32_Q15 +static inline spx_word32_t MULT16_32_Q15(spx_word16_t x, spx_word32_t y) { + int res; + int dummy; + __asm__ ( + "smull %0,%1,%2,%3 \n\t" + "mov %0, %0, lsr #15 \n\t" + "add %0, %0, %1, lsl #17 \n\t" + : "=&r"(res), "=&r" (dummy) + : "r"(y),"r"((int)x)); + return(res); +} + +#undef DIV32_16 +static inline short DIV32_16(int a, int b) +{ + int res=0; + int dead1, dead2, dead3, dead4, dead5; + __asm__ __volatile__ ( + "\teor %5, %0, %1\n" + "\tmovs %4, %0\n" + "\trsbmi %0, %0, #0 \n" + "\tmovs %4, %1\n" + "\trsbmi %1, %1, #0 \n" + "\tmov %4, #1\n" + + "\tsubs %3, %0, %1, asl #14 \n" + "\tmovpl %0, %3 \n" + "\torrpl %2, %2, %4, asl #14 \n" + + "\tsubs %3, %0, %1, asl #13 \n" + "\tmovpl %0, %3 \n" + "\torrpl %2, %2, %4, asl #13 \n" + + "\tsubs %3, %0, %1, asl #12 \n" + "\tmovpl %0, %3 \n" + "\torrpl %2, %2, %4, asl #12 \n" + + "\tsubs %3, %0, %1, asl #11 \n" + "\tmovpl %0, %3 \n" + "\torrpl %2, %2, %4, asl #11 \n" + + "\tsubs %3, %0, %1, asl #10 \n" + "\tmovpl %0, %3 \n" + "\torrpl %2, %2, %4, asl #10 \n" + + "\tsubs %3, %0, %1, asl #9 \n" + "\tmovpl %0, %3 \n" + "\torrpl %2, %2, %4, asl #9 \n" + + "\tsubs %3, %0, %1, asl #8 \n" + "\tmovpl %0, %3 \n" + "\torrpl %2, %2, %4, asl #8 \n" + + "\tsubs %3, %0, %1, asl #7 \n" + "\tmovpl %0, %3 \n" + "\torrpl %2, %2, %4, asl #7 \n" + + "\tsubs %3, %0, %1, asl #6 \n" + "\tmovpl %0, %3 \n" + "\torrpl %2, %2, %4, asl #6 \n" + + "\tsubs %3, %0, %1, asl #5 \n" + "\tmovpl %0, %3 \n" + "\torrpl %2, %2, %4, asl #5 \n" + + "\tsubs %3, %0, %1, asl #4 \n" + "\tmovpl %0, %3 \n" + "\torrpl %2, %2, %4, asl #4 \n" + + "\tsubs %3, %0, %1, asl #3 \n" + "\tmovpl %0, %3 \n" + "\torrpl %2, %2, %4, asl #3 \n" + + "\tsubs %3, %0, %1, asl #2 \n" + "\tmovpl %0, %3 \n" + "\torrpl %2, %2, %4, asl #2 \n" + + "\tsubs %3, %0, %1, asl #1 \n" + "\tmovpl %0, %3 \n" + "\torrpl %2, %2, %4, asl #1 \n" + + "\tsubs %3, %0, %1 \n" + "\tmovpl %0, %3 \n" + "\torrpl %2, %2, %4 \n" + + "\tmovs %5, %5, lsr #31 \n" + "\trsbne %2, %2, #0 \n" + : "=r" (dead1), "=r" (dead2), "=r" (res), + "=r" (dead3), "=r" (dead4), "=r" (dead5) + : "0" (a), "1" (b), "2" (res) + : "cc" + ); + return res; +} + + +#endif diff --git a/audio/deps/resampler/fixed_arm5e.h b/audio/deps/resampler/fixed_arm5e.h new file mode 100644 index 0000000..9fb11fc --- /dev/null +++ b/audio/deps/resampler/fixed_arm5e.h @@ -0,0 +1,160 @@ +/* Copyright (C) 2003 Jean-Marc Valin */ +/** + @file fixed_arm5e.h + @brief ARM-tuned fixed-point operations +*/ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + - Neither the name of the Xiph.org Foundation nor the names of its + contributors may be used to endorse or promote products derived from + this software without specific prior written permission. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR + CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef FIXED_ARM5E_H +#define FIXED_ARM5E_H + +#undef MULT16_16 +static inline spx_word32_t MULT16_16(spx_word16_t x, spx_word16_t y) { + int res; + __asm__ ("smulbb %0,%1,%2;\n" + : "=&r"(res) + : "%r"(x),"r"(y)); + return(res); +} + +#undef MAC16_16 +static inline spx_word32_t MAC16_16(spx_word32_t a, spx_word16_t x, spx_word32_t y) { + int res; + __asm__ ("smlabb %0,%1,%2,%3;\n" + : "=&r"(res) + : "%r"(x),"r"(y),"r"(a)); + return(res); +} + +#undef MULT16_32_Q15 +static inline spx_word32_t MULT16_32_Q15(spx_word16_t x, spx_word32_t y) { + int res; + __asm__ ("smulwb %0,%1,%2;\n" + : "=&r"(res) + : "%r"(y<<1),"r"(x)); + return(res); +} + +#undef MAC16_32_Q15 +static inline spx_word32_t MAC16_32_Q15(spx_word32_t a, spx_word16_t x, spx_word32_t y) { + int res; + __asm__ ("smlawb %0,%1,%2,%3;\n" + : "=&r"(res) + : "%r"(y<<1),"r"(x),"r"(a)); + return(res); +} + +#undef DIV32_16 +static inline short DIV32_16(int a, int b) +{ + int res=0; + int dead1, dead2, dead3, dead4, dead5; + __asm__ __volatile__ ( + "\teor %5, %0, %1\n" + "\tmovs %4, %0\n" + "\trsbmi %0, %0, #0 \n" + "\tmovs %4, %1\n" + "\trsbmi %1, %1, #0 \n" + "\tmov %4, #1\n" + + "\tsubs %3, %0, %1, asl #14 \n" + "\torrpl %2, %2, %4, asl #14 \n" + "\tmovpl %0, %3 \n" + + "\tsubs %3, %0, %1, asl #13 \n" + "\torrpl %2, %2, %4, asl #13 \n" + "\tmovpl %0, %3 \n" + + "\tsubs %3, %0, %1, asl #12 \n" + "\torrpl %2, %2, %4, asl #12 \n" + "\tmovpl %0, %3 \n" + + "\tsubs %3, %0, %1, asl #11 \n" + "\torrpl %2, %2, %4, asl #11 \n" + "\tmovpl %0, %3 \n" + + "\tsubs %3, %0, %1, asl #10 \n" + "\torrpl %2, %2, %4, asl #10 \n" + "\tmovpl %0, %3 \n" + + "\tsubs %3, %0, %1, asl #9 \n" + "\torrpl %2, %2, %4, asl #9 \n" + "\tmovpl %0, %3 \n" + + "\tsubs %3, %0, %1, asl #8 \n" + "\torrpl %2, %2, %4, asl #8 \n" + "\tmovpl %0, %3 \n" + + "\tsubs %3, %0, %1, asl #7 \n" + "\torrpl %2, %2, %4, asl #7 \n" + "\tmovpl %0, %3 \n" + + "\tsubs %3, %0, %1, asl #6 \n" + "\torrpl %2, %2, %4, asl #6 \n" + "\tmovpl %0, %3 \n" + + "\tsubs %3, %0, %1, asl #5 \n" + "\torrpl %2, %2, %4, asl #5 \n" + "\tmovpl %0, %3 \n" + + "\tsubs %3, %0, %1, asl #4 \n" + "\torrpl %2, %2, %4, asl #4 \n" + "\tmovpl %0, %3 \n" + + "\tsubs %3, %0, %1, asl #3 \n" + "\torrpl %2, %2, %4, asl #3 \n" + "\tmovpl %0, %3 \n" + + "\tsubs %3, %0, %1, asl #2 \n" + "\torrpl %2, %2, %4, asl #2 \n" + "\tmovpl %0, %3 \n" + + "\tsubs %3, %0, %1, asl #1 \n" + "\torrpl %2, %2, %4, asl #1 \n" + "\tmovpl %0, %3 \n" + + "\tsubs %3, %0, %1 \n" + "\torrpl %2, %2, %4 \n" + "\tmovpl %0, %3 \n" + + "\tmovs %5, %5, lsr #31 \n" + "\trsbne %2, %2, #0 \n" + : "=r" (dead1), "=r" (dead2), "=r" (res), + "=r" (dead3), "=r" (dead4), "=r" (dead5) + : "0" (a), "1" (b), "2" (res) + : "memory", "cc" + ); + return res; +} + + + + +#endif diff --git a/audio/deps/resampler/fixed_debug.h b/audio/deps/resampler/fixed_debug.h new file mode 100644 index 0000000..dbf02f1 --- /dev/null +++ b/audio/deps/resampler/fixed_debug.h @@ -0,0 +1,497 @@ +/* Copyright (C) 2003 Jean-Marc Valin */ +/** + @file fixed_debug.h + @brief Fixed-point operations with debugging +*/ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + - Neither the name of the Xiph.org Foundation nor the names of its + contributors may be used to endorse or promote products derived from + this software without specific prior written permission. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR + CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef FIXED_DEBUG_H +#define FIXED_DEBUG_H + +#include + +extern long long spx_mips; +#define MIPS_INC spx_mips++, + +#define QCONST16(x,bits) ((spx_word16_t)(.5+(x)*(((spx_word32_t)1)<<(bits)))) +#define QCONST32(x,bits) ((spx_word32_t)(.5+(x)*(((spx_word32_t)1)<<(bits)))) + + +#define VERIFY_SHORT(x) ((x)<=32767&&(x)>=-32768) +#define VERIFY_INT(x) ((x)<=2147483647LL&&(x)>=-2147483648LL) + +static inline short NEG16(int x) +{ + int res; + if (!VERIFY_SHORT(x)) + { + fprintf (stderr, "NEG16: input is not short: %d\n", (int)x); + } + res = -x; + if (!VERIFY_SHORT(res)) + fprintf (stderr, "NEG16: output is not short: %d\n", (int)res); + spx_mips++; + return res; +} +static inline int NEG32(long long x) +{ + long long res; + if (!VERIFY_INT(x)) + { + fprintf (stderr, "NEG16: input is not int: %d\n", (int)x); + } + res = -x; + if (!VERIFY_INT(res)) + fprintf (stderr, "NEG16: output is not int: %d\n", (int)res); + spx_mips++; + return res; +} + +#define EXTRACT16(x) _EXTRACT16(x, __FILE__, __LINE__) +static inline short _EXTRACT16(int x, char *file, int line) +{ + int res; + if (!VERIFY_SHORT(x)) + { + fprintf (stderr, "EXTRACT16: input is not short: %d in %s: line %d\n", x, file, line); + } + res = x; + spx_mips++; + return res; +} + +#define EXTEND32(x) _EXTEND32(x, __FILE__, __LINE__) +static inline int _EXTEND32(int x, char *file, int line) +{ + int res; + if (!VERIFY_SHORT(x)) + { + fprintf (stderr, "EXTEND32: input is not short: %d in %s: line %d\n", x, file, line); + } + res = x; + spx_mips++; + return res; +} + +#define SHR16(a, shift) _SHR16(a, shift, __FILE__, __LINE__) +static inline short _SHR16(int a, int shift, char *file, int line) +{ + int res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(shift)) + { + fprintf (stderr, "SHR16: inputs are not short: %d >> %d in %s: line %d\n", a, shift, file, line); + } + res = a>>shift; + if (!VERIFY_SHORT(res)) + fprintf (stderr, "SHR16: output is not short: %d in %s: line %d\n", res, file, line); + spx_mips++; + return res; +} +#define SHL16(a, shift) _SHL16(a, shift, __FILE__, __LINE__) +static inline short _SHL16(int a, int shift, char *file, int line) +{ + int res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(shift)) + { + fprintf (stderr, "SHL16: inputs are not short: %d %d in %s: line %d\n", a, shift, file, line); + } + res = a<>shift; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "SHR32: output is not int: %d\n", (int)res); + } + spx_mips++; + return res; +} +static inline int SHL32(long long a, int shift) +{ + long long res; + if (!VERIFY_INT(a) || !VERIFY_SHORT(shift)) + { + fprintf (stderr, "SHL32: inputs are not int: %d %d\n", (int)a, shift); + } + res = a<>1))),shift)) +#define PSHR32(a,shift) (SHR32(ADD32((a),((EXTEND32(1)<<((shift))>>1))),shift)) +#define VSHR32(a, shift) (((shift)>0) ? SHR32(a, shift) : SHL32(a, -(shift))) + +#define SATURATE16(x,a) (((x)>(a) ? (a) : (x)<-(a) ? -(a) : (x))) +#define SATURATE32(x,a) (((x)>(a) ? (a) : (x)<-(a) ? -(a) : (x))) + +//#define SHR(a,shift) ((a) >> (shift)) +//#define SHL(a,shift) ((a) << (shift)) + +#define ADD16(a, b) _ADD16(a, b, __FILE__, __LINE__) +static inline short _ADD16(int a, int b, char *file, int line) +{ + int res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "ADD16: inputs are not short: %d %d in %s: line %d\n", a, b, file, line); + } + res = a+b; + if (!VERIFY_SHORT(res)) + { + fprintf (stderr, "ADD16: output is not short: %d+%d=%d in %s: line %d\n", a,b,res, file, line); + } + spx_mips++; + return res; +} + +#define SUB16(a, b) _SUB16(a, b, __FILE__, __LINE__) +static inline short _SUB16(int a, int b, char *file, int line) +{ + int res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "SUB16: inputs are not short: %d %d in %s: line %d\n", a, b, file, line); + } + res = a-b; + if (!VERIFY_SHORT(res)) + fprintf (stderr, "SUB16: output is not short: %d in %s: line %d\n", res, file, line); + spx_mips++; + return res; +} + +#define ADD32(a, b) _ADD32(a, b, __FILE__, __LINE__) +static inline int _ADD32(long long a, long long b, char *file, int line) +{ + long long res; + if (!VERIFY_INT(a) || !VERIFY_INT(b)) + { + fprintf (stderr, "ADD32: inputs are not int: %d %d in %s: line %d\n", (int)a, (int)b, file, line); + } + res = a+b; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "ADD32: output is not int: %d in %s: line %d\n", (int)res, file, line); + } + spx_mips++; + return res; +} + +static inline int SUB32(long long a, long long b) +{ + long long res; + if (!VERIFY_INT(a) || !VERIFY_INT(b)) + { + fprintf (stderr, "SUB32: inputs are not int: %d %d\n", (int)a, (int)b); + } + res = a-b; + if (!VERIFY_INT(res)) + fprintf (stderr, "SUB32: output is not int: %d\n", (int)res); + spx_mips++; + return res; +} + +#define ADD64(a,b) (MIPS_INC(a)+(b)) + +/* result fits in 16 bits */ +static inline short MULT16_16_16(int a, int b) +{ + int res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "MULT16_16_16: inputs are not short: %d %d\n", a, b); + } + res = a*b; + if (!VERIFY_SHORT(res)) + fprintf (stderr, "MULT16_16_16: output is not short: %d\n", res); + spx_mips++; + return res; +} + +/* result fits in 32 bits */ +static inline int MULT16_32_32(int a, long long b) +{ + long long res; + if (!VERIFY_SHORT(a) || !VERIFY_INT(b)) + { + fprintf (stderr, "MULT16_32_32: inputs are not short+int: %d %d\n", a, (int)b); + } + res = a*b; + if (!VERIFY_INT(res)) + fprintf (stderr, "MULT16_32_32: output is not int: %d\n", (int)res); + spx_mips++; + return res; +} + +#define MULT16_16(a, b) _MULT16_16(a, b, __FILE__, __LINE__) +static inline int _MULT16_16(int a, int b, char *file, int line) +{ + long long res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "MULT16_16: inputs are not short: %d %d in %s: line %d\n", a, b, file, line); + } + res = ((long long)a)*b; + if (!VERIFY_INT(res)) + fprintf (stderr, "MULT16_16: output is not int: %d in %s: line %d\n", (int)res, file, line); + spx_mips++; + return res; +} + +#define MAC16_16(c,a,b) (spx_mips--,ADD32((c),MULT16_16((a),(b)))) +#define MAC16_16_Q11(c,a,b) (EXTRACT16(ADD16((c),EXTRACT16(SHR32(MULT16_16((a),(b)),11))))) +#define MAC16_16_Q13(c,a,b) (EXTRACT16(ADD16((c),EXTRACT16(SHR32(MULT16_16((a),(b)),13))))) +#define MAC16_16_P13(c,a,b) (EXTRACT16(ADD32((c),SHR32(ADD32(4096,MULT16_16((a),(b))),13)))) + + +#define MULT16_32_QX(a, b, Q) _MULT16_32_QX(a, b, Q, __FILE__, __LINE__) +static inline int _MULT16_32_QX(int a, long long b, int Q, char *file, int line) +{ + long long res; + if (!VERIFY_SHORT(a) || !VERIFY_INT(b)) + { + fprintf (stderr, "MULT16_32_Q%d: inputs are not short+int: %d %d in %s: line %d\n", Q, (int)a, (int)b, file, line); + } + if (ABS(b)>>(16+Q)) + fprintf (stderr, "MULT16_32_Q%d: second operand too large: %d %d in %s: line %d\n", Q, (int)a, (int)b, file, line); + res = (((long long)a)*(long long)b) >> Q; + if (!VERIFY_INT(res)) + fprintf (stderr, "MULT16_32_Q%d: output is not int: %d*%d=%d in %s: line %d\n", Q, (int)a, (int)b,(int)res, file, line); + spx_mips+=5; + return res; +} + +static inline int MULT16_32_PX(int a, long long b, int Q) +{ + long long res; + if (!VERIFY_SHORT(a) || !VERIFY_INT(b)) + { + fprintf (stderr, "MULT16_32_P%d: inputs are not short+int: %d %d\n", Q, (int)a, (int)b); + } + if (ABS(b)>>(16+Q)) + fprintf (stderr, "MULT16_32_Q%d: second operand too large: %d %d\n", Q, (int)a, (int)b); + res = ((((long long)a)*(long long)b) + ((EXTEND32(1)<>1))>> Q; + if (!VERIFY_INT(res)) + fprintf (stderr, "MULT16_32_P%d: output is not int: %d*%d=%d\n", Q, (int)a, (int)b,(int)res); + spx_mips+=5; + return res; +} + + +#define MULT16_32_Q15(a,b) MULT16_32_QX(a,b,15) +#define MULT16_32_P15(a,b) MULT16_32_PX(a,b,15) +#define MAC16_32_Q15(c,a,b) ADD32((c),MULT16_32_Q15((a),(b))) + +static inline int SATURATE(int a, int b) +{ + if (a>b) + a=b; + if (a<-b) + a = -b; + return a; +} + +static inline int MULT16_16_Q11_32(int a, int b) +{ + long long res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "MULT16_16_Q11: inputs are not short: %d %d\n", a, b); + } + res = ((long long)a)*b; + res >>= 11; + if (!VERIFY_INT(res)) + fprintf (stderr, "MULT16_16_Q11: output is not short: %d*%d=%d\n", (int)a, (int)b, (int)res); + spx_mips+=3; + return res; +} +static inline short MULT16_16_Q13(int a, int b) +{ + long long res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "MULT16_16_Q13: inputs are not short: %d %d\n", a, b); + } + res = ((long long)a)*b; + res >>= 13; + if (!VERIFY_SHORT(res)) + fprintf (stderr, "MULT16_16_Q13: output is not short: %d*%d=%d\n", a, b, (int)res); + spx_mips+=3; + return res; +} +static inline short MULT16_16_Q14(int a, int b) +{ + long long res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "MULT16_16_Q14: inputs are not short: %d %d\n", a, b); + } + res = ((long long)a)*b; + res >>= 14; + if (!VERIFY_SHORT(res)) + fprintf (stderr, "MULT16_16_Q14: output is not short: %d\n", (int)res); + spx_mips+=3; + return res; +} +static inline short MULT16_16_Q15(int a, int b) +{ + long long res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "MULT16_16_Q15: inputs are not short: %d %d\n", a, b); + } + res = ((long long)a)*b; + res >>= 15; + if (!VERIFY_SHORT(res)) + { + fprintf (stderr, "MULT16_16_Q15: output is not short: %d\n", (int)res); + } + spx_mips+=3; + return res; +} + +static inline short MULT16_16_P13(int a, int b) +{ + long long res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "MULT16_16_P13: inputs are not short: %d %d\n", a, b); + } + res = ((long long)a)*b; + res += 4096; + if (!VERIFY_INT(res)) + fprintf (stderr, "MULT16_16_P13: overflow: %d*%d=%d\n", a, b, (int)res); + res >>= 13; + if (!VERIFY_SHORT(res)) + fprintf (stderr, "MULT16_16_P13: output is not short: %d*%d=%d\n", a, b, (int)res); + spx_mips+=4; + return res; +} +static inline short MULT16_16_P14(int a, int b) +{ + long long res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "MULT16_16_P14: inputs are not short: %d %d\n", a, b); + } + res = ((long long)a)*b; + res += 8192; + if (!VERIFY_INT(res)) + fprintf (stderr, "MULT16_16_P14: overflow: %d*%d=%d\n", a, b, (int)res); + res >>= 14; + if (!VERIFY_SHORT(res)) + fprintf (stderr, "MULT16_16_P14: output is not short: %d*%d=%d\n", a, b, (int)res); + spx_mips+=4; + return res; +} +static inline short MULT16_16_P15(int a, int b) +{ + long long res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "MULT16_16_P15: inputs are not short: %d %d\n", a, b); + } + res = ((long long)a)*b; + res += 16384; + if (!VERIFY_INT(res)) + fprintf (stderr, "MULT16_16_P15: overflow: %d*%d=%d\n", a, b, (int)res); + res >>= 15; + if (!VERIFY_SHORT(res)) + fprintf (stderr, "MULT16_16_P15: output is not short: %d*%d=%d\n", a, b, (int)res); + spx_mips+=4; + return res; +} + +#define DIV32_16(a, b) _DIV32_16(a, b, __FILE__, __LINE__) + +static inline int _DIV32_16(long long a, long long b, char *file, int line) +{ + long long res; + if (b==0) + { + fprintf(stderr, "DIV32_16: divide by zero: %d/%d in %s: line %d\n", (int)a, (int)b, file, line); + return 0; + } + if (!VERIFY_INT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "DIV32_16: inputs are not int/short: %d %d in %s: line %d\n", (int)a, (int)b, file, line); + } + res = a/b; + if (!VERIFY_SHORT(res)) + { + fprintf (stderr, "DIV32_16: output is not short: %d / %d = %d in %s: line %d\n", (int)a,(int)b,(int)res, file, line); + if (res>32767) + res = 32767; + if (res<-32768) + res = -32768; + } + spx_mips+=20; + return res; +} + +#define DIV32(a, b) _DIV32(a, b, __FILE__, __LINE__) +static inline int _DIV32(long long a, long long b, char *file, int line) +{ + long long res; + if (b==0) + { + fprintf(stderr, "DIV32: divide by zero: %d/%d in %s: line %d\n", (int)a, (int)b, file, line); + return 0; + } + + if (!VERIFY_INT(a) || !VERIFY_INT(b)) + { + fprintf (stderr, "DIV32: inputs are not int/short: %d %d in %s: line %d\n", (int)a, (int)b, file, line); + } + res = a/b; + if (!VERIFY_INT(res)) + fprintf (stderr, "DIV32: output is not int: %d in %s: line %d\n", (int)res, file, line); + spx_mips+=36; + return res; +} +#define PDIV32(a,b) DIV32(ADD32((a),(b)>>1),b) +#define PDIV32_16(a,b) DIV32_16(ADD32((a),(b)>>1),b) + +#endif diff --git a/audio/deps/resampler/fixed_generic.h b/audio/deps/resampler/fixed_generic.h new file mode 100644 index 0000000..09366c3 --- /dev/null +++ b/audio/deps/resampler/fixed_generic.h @@ -0,0 +1,106 @@ +/* Copyright (C) 2003 Jean-Marc Valin */ +/** + @file fixed_generic.h + @brief Generic fixed-point operations +*/ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + - Neither the name of the Xiph.org Foundation nor the names of its + contributors may be used to endorse or promote products derived from + this software without specific prior written permission. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR + CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef FIXED_GENERIC_H +#define FIXED_GENERIC_H + +#define QCONST16(x,bits) ((spx_word16_t)(.5+(x)*(((spx_word32_t)1)<<(bits)))) +#define QCONST32(x,bits) ((spx_word32_t)(.5+(x)*(((spx_word32_t)1)<<(bits)))) + +#define NEG16(x) (-(x)) +#define NEG32(x) (-(x)) +#define EXTRACT16(x) ((spx_word16_t)(x)) +#define EXTEND32(x) ((spx_word32_t)(x)) +#define SHR16(a,shift) ((a) >> (shift)) +#define SHL16(a,shift) ((a) << (shift)) +#define SHR32(a,shift) ((a) >> (shift)) +#define SHL32(a,shift) ((a) << (shift)) +#define PSHR16(a,shift) (SHR16((a)+((1<<((shift))>>1)),shift)) +#define PSHR32(a,shift) (SHR32((a)+((EXTEND32(1)<<((shift))>>1)),shift)) +#define VSHR32(a, shift) (((shift)>0) ? SHR32(a, shift) : SHL32(a, -(shift))) +#define SATURATE16(x,a) (((x)>(a) ? (a) : (x)<-(a) ? -(a) : (x))) +#define SATURATE32(x,a) (((x)>(a) ? (a) : (x)<-(a) ? -(a) : (x))) + +#define SATURATE32PSHR(x,shift,a) (((x)>=(SHL32(a,shift))) ? (a) : \ + (x)<=-(SHL32(a,shift)) ? -(a) : \ + (PSHR32(x, shift))) + +#define SHR(a,shift) ((a) >> (shift)) +#define SHL(a,shift) ((spx_word32_t)(a) << (shift)) +#define PSHR(a,shift) (SHR((a)+((EXTEND32(1)<<((shift))>>1)),shift)) +#define SATURATE(x,a) (((x)>(a) ? (a) : (x)<-(a) ? -(a) : (x))) + + +#define ADD16(a,b) ((spx_word16_t)((spx_word16_t)(a)+(spx_word16_t)(b))) +#define SUB16(a,b) ((spx_word16_t)(a)-(spx_word16_t)(b)) +#define ADD32(a,b) ((spx_word32_t)(a)+(spx_word32_t)(b)) +#define SUB32(a,b) ((spx_word32_t)(a)-(spx_word32_t)(b)) + + +/* result fits in 16 bits */ +#define MULT16_16_16(a,b) (((spx_word16_t)(a))*((spx_word16_t)(b))) +/* result fits in 32 bits */ +#define MULT16_32_32(a,b) (((spx_word16_t)(a))*((spx_word32_t)(b))) + +/* (spx_word32_t)(spx_word16_t) gives TI compiler a hint that it's 16x16->32 multiply */ +#define MULT16_16(a,b) (((spx_word32_t)(spx_word16_t)(a))*((spx_word32_t)(spx_word16_t)(b))) + +#define MAC16_16(c,a,b) (ADD32((c),MULT16_16((a),(b)))) + +#define MULT16_32_P15(a,b) ADD32(MULT16_32_32(a,SHR((b),15)), PSHR(MULT16_16((a),((b)&0x00007fff)),15)) +#define MULT16_32_Q15(a,b) ADD32(MULT16_32_32(a,SHR((b),15)), SHR(MULT16_16((a),((b)&0x00007fff)),15)) +#define MAC16_32_Q15(c,a,b) ADD32(c,MULT16_32_Q15(a,b)) + + +#define MAC16_16_Q11(c,a,b) (ADD32((c),SHR(MULT16_16((a),(b)),11))) +#define MAC16_16_Q13(c,a,b) (ADD32((c),SHR(MULT16_16((a),(b)),13))) +#define MAC16_16_P13(c,a,b) (ADD32((c),SHR(ADD32(4096,MULT16_16((a),(b))),13))) + +#define MULT16_16_Q11_32(a,b) (SHR(MULT16_16((a),(b)),11)) +#define MULT16_16_Q13(a,b) (SHR(MULT16_16((a),(b)),13)) +#define MULT16_16_Q14(a,b) (SHR(MULT16_16((a),(b)),14)) +#define MULT16_16_Q15(a,b) (SHR(MULT16_16((a),(b)),15)) + +#define MULT16_16_P13(a,b) (SHR(ADD32(4096,MULT16_16((a),(b))),13)) +#define MULT16_16_P14(a,b) (SHR(ADD32(8192,MULT16_16((a),(b))),14)) +#define MULT16_16_P15(a,b) (SHR(ADD32(16384,MULT16_16((a),(b))),15)) + +#define MUL_16_32_R15(a,bh,bl) ADD32(MULT16_16((a),(bh)), SHR(MULT16_16((a),(bl)),15)) + +#define DIV32_16(a,b) ((spx_word16_t)(((spx_word32_t)(a))/((spx_word16_t)(b)))) +#define PDIV32_16(a,b) ((spx_word16_t)(((spx_word32_t)(a)+((spx_word16_t)(b)>>1))/((spx_word16_t)(b)))) +#define DIV32(a,b) (((spx_word32_t)(a))/((spx_word32_t)(b))) +#define PDIV32(a,b) (((spx_word32_t)(a)+((spx_word16_t)(b)>>1))/((spx_word32_t)(b))) + +#endif diff --git a/audio/deps/resampler/os_support.h b/audio/deps/resampler/os_support.h new file mode 100644 index 0000000..4e16aea --- /dev/null +++ b/audio/deps/resampler/os_support.h @@ -0,0 +1,89 @@ +/* Copyright (C) 2007 Jean-Marc Valin + + File: os_support.h + This is the (tiny) OS abstraction layer. Aside from math.h, this is the + only place where system headers are allowed. + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions are + met: + + 1. Redistributions of source code must retain the above copyright notice, + this list of conditions and the following disclaimer. + + 2. Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + 3. The name of the author may not be used to endorse or promote products + derived from this software without specific prior written permission. + + THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR + IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES + OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + DISCLAIMED. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, + INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR + SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, + STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN + ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE + POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef OS_SUPPORT_H +#define OS_SUPPORT_H + +#include +#include +#include + +#include "config.h" + +#ifdef OS_SUPPORT_CUSTOM +#include "os_support_custom.h" +#endif + +/** Speex wrapper for calloc. To do your own dynamic allocation, all you need to do is replace this function, speex_realloc and speex_free + NOTE: speex_alloc needs to CLEAR THE MEMORY */ +#ifndef OVERRIDE_SPEEX_ALLOC +static inline void *speex_alloc (int size) +{ + /* WARNING: this is not equivalent to malloc(). If you want to use malloc() + or your own allocator, YOU NEED TO CLEAR THE MEMORY ALLOCATED. Otherwise + you will experience strange bugs */ + return calloc(size,1); +} +#endif + +/** Speex wrapper for realloc. To do your own dynamic allocation, all you need to do is replace this function, speex_alloc and speex_free */ +#ifndef OVERRIDE_SPEEX_REALLOC +static inline void *speex_realloc (void *ptr, int size) +{ + return realloc(ptr, size); +} +#endif + +/** Speex wrapper for calloc. To do your own dynamic allocation, all you need to do is replace this function, speex_realloc and speex_alloc */ +#ifndef OVERRIDE_SPEEX_FREE +static inline void speex_free (void *ptr) +{ + free(ptr); +} +#endif + +#ifndef OVERRIDE_SPEEX_WARNING +static inline void speex_warning(const char *str) +{ +#ifndef DISABLE_WARNINGS + fprintf (stderr, "warning: %s\n", str); +#endif +} +#endif + + + + + +#endif + diff --git a/audio/deps/resampler/resample.c b/audio/deps/resampler/resample.c new file mode 100644 index 0000000..987f9e3 --- /dev/null +++ b/audio/deps/resampler/resample.c @@ -0,0 +1,1051 @@ +/* Copyright (C) 2007-2008 Jean-Marc Valin + Copyright (C) 2008 Thorvald Natvig + + File: resample.c + Arbitrary resampling code + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions are + met: + + 1. Redistributions of source code must retain the above copyright notice, + this list of conditions and the following disclaimer. + + 2. Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + 3. The name of the author may not be used to endorse or promote products + derived from this software without specific prior written permission. + + THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR + IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES + OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + DISCLAIMED. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, + INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR + SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, + STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN + ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE + POSSIBILITY OF SUCH DAMAGE. +*/ + +/* + The design goals of this code are: + - Very fast algorithm + - SIMD-friendly algorithm + - Low memory requirement + - Good *perceptual* quality (and not best SNR) + + Warning: This resampler is relatively new. Although I think I got rid of + all the major bugs and I don't expect the API to change anymore, there + may be something I've missed. So use with caution. + + This algorithm is based on this original resampling algorithm: + Smith, Julius O. Digital Audio Resampling Home Page + Center for Computer Research in Music and Acoustics (CCRMA), + Stanford University, 2007. + Web published at https://ccrma.stanford.edu/~jos/resample/. + + There is one main difference, though. This resampler uses cubic + interpolation instead of linear interpolation in the above paper. This + makes the table much smaller and makes it possible to compute that table + on a per-stream basis. In turn, being able to tweak the table for each + stream makes it possible to both reduce complexity on simple ratios + (e.g. 2/3), and get rid of the rounding operations in the inner loop. + The latter both reduces CPU time and makes the algorithm more SIMD-friendly. +*/ + +#include "config.h" + +#include +#include "speex_resampler.h" +#include "arch.h" +#include "os_support.h" + +#include +#include + +#ifndef M_PI +#define M_PI 3.14159265358979323846 +#endif + +#define IMAX(a,b) ((a) > (b) ? (a) : (b)) +#define IMIN(a,b) ((a) < (b) ? (a) : (b)) + +#ifdef USE_NEON +#include "resample_neon.h" +#endif + +/* Numer of elements to allocate on the stack */ +#ifdef VAR_ARRAYS +#define FIXED_STACK_ALLOC 8192 +#else +#define FIXED_STACK_ALLOC 1024 +#endif + +typedef int (*resampler_basic_func)(SpeexResamplerState *, spx_uint32_t , const spx_word16_t *, spx_uint32_t *, spx_word16_t *, spx_uint32_t *); + +/** Set (change) the input/output sampling rates and resampling ratio + * (fractional values in Hz supported). + * @param st Resampler state + * @param ratio_num Numerator of the sampling rate ratio + * @param ratio_den Denominator of the sampling rate ratio + * @param in_rate Input sampling rate rounded to the nearest integer (in Hz). + * @param out_rate Output sampling rate rounded to the nearest integer (in Hz). + */ +int speex_resampler_set_rate_frac(SpeexResamplerState *st, + spx_uint32_t ratio_num, + spx_uint32_t ratio_den, + spx_uint32_t in_rate, + spx_uint32_t out_rate); + +/** Create a new resampler with fractional input/output rates. The sampling + * rate ratio is an arbitrary rational number with both the numerator and + * denominator being 32-bit integers. + * @param nb_channels Number of channels to be processed + * @param ratio_num Numerator of the sampling rate ratio + * @param ratio_den Denominator of the sampling rate ratio + * @param in_rate Input sampling rate rounded to the nearest integer (in Hz). + * @param out_rate Output sampling rate rounded to the nearest integer (in Hz). + * @param quality Resampling quality between 0 and 10, where 0 has poor quality + * and 10 has very high quality. + * @return Newly created resampler state + * @retval NULL Error: not enough memory + */ +SpeexResamplerState *speex_resampler_init_frac(spx_uint32_t nb_channels, + spx_uint32_t ratio_num, + spx_uint32_t ratio_den, + spx_uint32_t in_rate, + spx_uint32_t out_rate, + int quality, + int *err); + +struct SpeexResamplerState_ { + spx_uint32_t in_rate; + spx_uint32_t out_rate; + spx_uint32_t num_rate; + spx_uint32_t den_rate; + + int quality; + spx_uint32_t nb_channels; + spx_uint32_t filt_len; + spx_uint32_t mem_alloc_size; + spx_uint32_t buffer_size; + int int_advance; + int frac_advance; + float cutoff; + spx_uint32_t oversample; + int initialised; + int started; + + /* These are per-channel */ + spx_int32_t *last_sample; + spx_uint32_t *samp_frac_num; + spx_uint32_t *magic_samples; + + spx_word16_t *mem; + spx_word16_t *sinc_table; + spx_uint32_t sinc_table_length; + resampler_basic_func resampler_ptr; + + int in_stride; + int out_stride; +} ; + +static const double kaiser12_table[68] = { + 0.99859849, 1.00000000, 0.99859849, 0.99440475, 0.98745105, 0.97779076, + 0.96549770, 0.95066529, 0.93340547, 0.91384741, 0.89213598, 0.86843014, + 0.84290116, 0.81573067, 0.78710866, 0.75723148, 0.72629970, 0.69451601, + 0.66208321, 0.62920216, 0.59606986, 0.56287762, 0.52980938, 0.49704014, + 0.46473455, 0.43304576, 0.40211431, 0.37206735, 0.34301800, 0.31506490, + 0.28829195, 0.26276832, 0.23854851, 0.21567274, 0.19416736, 0.17404546, + 0.15530766, 0.13794294, 0.12192957, 0.10723616, 0.09382272, 0.08164178, + 0.07063950, 0.06075685, 0.05193064, 0.04409466, 0.03718069, 0.03111947, + 0.02584161, 0.02127838, 0.01736250, 0.01402878, 0.01121463, 0.00886058, + 0.00691064, 0.00531256, 0.00401805, 0.00298291, 0.00216702, 0.00153438, + 0.00105297, 0.00069463, 0.00043489, 0.00025272, 0.00013031, 0.0000527734, + 0.00001000, 0.00000000}; +/* +static const double kaiser12_table[36] = { + 0.99440475, 1.00000000, 0.99440475, 0.97779076, 0.95066529, 0.91384741, + 0.86843014, 0.81573067, 0.75723148, 0.69451601, 0.62920216, 0.56287762, + 0.49704014, 0.43304576, 0.37206735, 0.31506490, 0.26276832, 0.21567274, + 0.17404546, 0.13794294, 0.10723616, 0.08164178, 0.06075685, 0.04409466, + 0.03111947, 0.02127838, 0.01402878, 0.00886058, 0.00531256, 0.00298291, + 0.00153438, 0.00069463, 0.00025272, 0.0000527734, 0.00000500, 0.00000000}; +*/ +static const double kaiser10_table[36] = { + 0.99537781, 1.00000000, 0.99537781, 0.98162644, 0.95908712, 0.92831446, + 0.89005583, 0.84522401, 0.79486424, 0.74011713, 0.68217934, 0.62226347, + 0.56155915, 0.50119680, 0.44221549, 0.38553619, 0.33194107, 0.28205962, + 0.23636152, 0.19515633, 0.15859932, 0.12670280, 0.09935205, 0.07632451, + 0.05731132, 0.04193980, 0.02979584, 0.02044510, 0.01345224, 0.00839739, + 0.00488951, 0.00257636, 0.00115101, 0.00035515, 0.00000000, 0.00000000}; + +static const double kaiser8_table[36] = { + 0.99635258, 1.00000000, 0.99635258, 0.98548012, 0.96759014, 0.94302200, + 0.91223751, 0.87580811, 0.83439927, 0.78875245, 0.73966538, 0.68797126, + 0.63451750, 0.58014482, 0.52566725, 0.47185369, 0.41941150, 0.36897272, + 0.32108304, 0.27619388, 0.23465776, 0.19672670, 0.16255380, 0.13219758, + 0.10562887, 0.08273982, 0.06335451, 0.04724088, 0.03412321, 0.02369490, + 0.01563093, 0.00959968, 0.00527363, 0.00233883, 0.00050000, 0.00000000}; + +static const double kaiser6_table[36] = { + 0.99733006, 1.00000000, 0.99733006, 0.98935595, 0.97618418, 0.95799003, + 0.93501423, 0.90755855, 0.87598009, 0.84068475, 0.80211977, 0.76076565, + 0.71712752, 0.67172623, 0.62508937, 0.57774224, 0.53019925, 0.48295561, + 0.43647969, 0.39120616, 0.34752997, 0.30580127, 0.26632152, 0.22934058, + 0.19505503, 0.16360756, 0.13508755, 0.10953262, 0.08693120, 0.06722600, + 0.05031820, 0.03607231, 0.02432151, 0.01487334, 0.00752000, 0.00000000}; + +struct FuncDef { + const double *table; + int oversample; +}; + +static const struct FuncDef kaiser12_funcdef = {kaiser12_table, 64}; +#define KAISER12 (&kaiser12_funcdef) +static const struct FuncDef kaiser10_funcdef = {kaiser10_table, 32}; +#define KAISER10 (&kaiser10_funcdef) +static const struct FuncDef kaiser8_funcdef = {kaiser8_table, 32}; +#define KAISER8 (&kaiser8_funcdef) +static const struct FuncDef kaiser6_funcdef = {kaiser6_table, 32}; +#define KAISER6 (&kaiser6_funcdef) + +struct QualityMapping { + int base_length; + int oversample; + float downsample_bandwidth; + float upsample_bandwidth; + const struct FuncDef *window_func; +}; + + +/* This table maps conversion quality to internal parameters. There are two + reasons that explain why the up-sampling bandwidth is larger than the + down-sampling bandwidth: + 1) When up-sampling, we can assume that the spectrum is already attenuated + close to the Nyquist rate (from an A/D or a previous resampling filter) + 2) Any aliasing that occurs very close to the Nyquist rate will be masked + by the sinusoids/noise just below the Nyquist rate (guaranteed only for + up-sampling). +*/ +static const struct QualityMapping quality_map[11] = { + { 8, 4, 0.830f, 0.860f, KAISER6 }, /* Q0 */ + { 16, 4, 0.850f, 0.880f, KAISER6 }, /* Q1 */ + { 32, 4, 0.882f, 0.910f, KAISER6 }, /* Q2 */ /* 82.3% cutoff ( ~60 dB stop) 6 */ + { 48, 8, 0.895f, 0.917f, KAISER8 }, /* Q3 */ /* 84.9% cutoff ( ~80 dB stop) 8 */ + { 64, 8, 0.921f, 0.940f, KAISER8 }, /* Q4 */ /* 88.7% cutoff ( ~80 dB stop) 8 */ + { 80, 16, 0.922f, 0.940f, KAISER10}, /* Q5 */ /* 89.1% cutoff (~100 dB stop) 10 */ + { 96, 16, 0.940f, 0.945f, KAISER10}, /* Q6 */ /* 91.5% cutoff (~100 dB stop) 10 */ + {128, 16, 0.950f, 0.950f, KAISER10}, /* Q7 */ /* 93.1% cutoff (~100 dB stop) 10 */ + {160, 16, 0.960f, 0.960f, KAISER10}, /* Q8 */ /* 94.5% cutoff (~100 dB stop) 10 */ + {192, 32, 0.968f, 0.968f, KAISER12}, /* Q9 */ /* 95.5% cutoff (~100 dB stop) 10 */ + {256, 32, 0.975f, 0.975f, KAISER12}, /* Q10 */ /* 96.6% cutoff (~100 dB stop) 10 */ +}; +/*8,24,40,56,80,104,128,160,200,256,320*/ +static double compute_func(float x, const struct FuncDef *func) +{ + float y, frac; + double interp[4]; + int ind; + y = x*func->oversample; + ind = (int)floor(y); + frac = (y-ind); + /* CSE with handle the repeated powers */ + interp[3] = -0.1666666667*frac + 0.1666666667*(frac*frac*frac); + interp[2] = frac + 0.5*(frac*frac) - 0.5*(frac*frac*frac); + /*interp[2] = 1.f - 0.5f*frac - frac*frac + 0.5f*frac*frac*frac;*/ + interp[0] = -0.3333333333*frac + 0.5*(frac*frac) - 0.1666666667*(frac*frac*frac); + /* Just to make sure we don't have rounding problems */ + interp[1] = 1.f-interp[3]-interp[2]-interp[0]; + + /*sum = frac*accum[1] + (1-frac)*accum[2];*/ + return interp[0]*func->table[ind] + interp[1]*func->table[ind+1] + interp[2]*func->table[ind+2] + interp[3]*func->table[ind+3]; +} + + +/* The slow way of computing a sinc for the table. Should improve that some day */ +static spx_word16_t sinc(float cutoff, float x, int N, const struct FuncDef *window_func) +{ + /*fprintf (stderr, "%f ", x);*/ + float xx = x * cutoff; + if (fabs(x)<1e-6f) + return WORD2INT(32768.*cutoff); + else if (fabs(x) > .5f*N) + return 0; + /*FIXME: Can it really be any slower than this? */ + return WORD2INT(32768.*cutoff*sin(M_PI*xx)/(M_PI*xx) * compute_func(fabs(2.*x/N), window_func)); +} + + +static void cubic_coef(spx_word16_t x, spx_word16_t interp[4]) +{ + /* Compute interpolation coefficients. I'm not sure whether this corresponds to cubic interpolation + but I know it's MMSE-optimal on a sinc */ + spx_word16_t x2, x3; + x2 = MULT16_16_P15(x, x); + x3 = MULT16_16_P15(x, x2); + interp[0] = PSHR32(MULT16_16(QCONST16(-0.16667f, 15),x) + MULT16_16(QCONST16(0.16667f, 15),x3),15); + interp[1] = EXTRACT16(EXTEND32(x) + SHR32(SUB32(EXTEND32(x2),EXTEND32(x3)),1)); + interp[3] = PSHR32(MULT16_16(QCONST16(-0.33333f, 15),x) + MULT16_16(QCONST16(.5f,15),x2) - MULT16_16(QCONST16(0.16667f, 15),x3),15); + /* Just to make sure we don't have rounding problems */ + interp[2] = Q15_ONE-interp[0]-interp[1]-interp[3]; + if (interp[2]<32767) + interp[2]+=1; +} + + +static int resampler_basic_direct_single(SpeexResamplerState *st, spx_uint32_t channel_index, const spx_word16_t *in, spx_uint32_t *in_len, spx_word16_t *out, spx_uint32_t *out_len) +{ + const int N = st->filt_len; + int out_sample = 0; + int last_sample = st->last_sample[channel_index]; + spx_uint32_t samp_frac_num = st->samp_frac_num[channel_index]; + const spx_word16_t *sinc_table = st->sinc_table; + const int out_stride = st->out_stride; + const int int_advance = st->int_advance; + const int frac_advance = st->frac_advance; + const spx_uint32_t den_rate = st->den_rate; + spx_word32_t sum; + + while (!(last_sample >= (spx_int32_t)*in_len || out_sample >= (spx_int32_t)*out_len)) + { + const spx_word16_t *sinct = & sinc_table[samp_frac_num*N]; + const spx_word16_t *iptr = & in[last_sample]; + +#ifndef OVERRIDE_INNER_PRODUCT_SINGLE + int j; + sum = 0; + for(j=0;j= den_rate) + { + samp_frac_num -= den_rate; + last_sample++; + } + } + + st->last_sample[channel_index] = last_sample; + st->samp_frac_num[channel_index] = samp_frac_num; + return out_sample; +} + + +static int resampler_basic_interpolate_single(SpeexResamplerState *st, spx_uint32_t channel_index, const spx_word16_t *in, spx_uint32_t *in_len, spx_word16_t *out, spx_uint32_t *out_len) +{ + const int N = st->filt_len; + int out_sample = 0; + int last_sample = st->last_sample[channel_index]; + spx_uint32_t samp_frac_num = st->samp_frac_num[channel_index]; + const int out_stride = st->out_stride; + const int int_advance = st->int_advance; + const int frac_advance = st->frac_advance; + const spx_uint32_t den_rate = st->den_rate; + spx_word32_t sum; + + while (!(last_sample >= (spx_int32_t)*in_len || out_sample >= (spx_int32_t)*out_len)) + { + const spx_word16_t *iptr = & in[last_sample]; + + const int offset = samp_frac_num*st->oversample/st->den_rate; + const spx_word16_t frac = PDIV32(SHL32((samp_frac_num*st->oversample) % st->den_rate,15),st->den_rate); + spx_word16_t interp[4]; + + int j; + spx_word32_t accum[4] = {0,0,0,0}; + + for(j=0;jsinc_table[4+(j+1)*st->oversample-offset-2]); + accum[1] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset-1]); + accum[2] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset]); + accum[3] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset+1]); + } + + cubic_coef(frac, interp); + sum = MULT16_32_Q15(interp[0],accum[0]) + MULT16_32_Q15(interp[1],accum[1]) + MULT16_32_Q15(interp[2],accum[2]) + MULT16_32_Q15(interp[3],accum[3]); + sum = SATURATE32PSHR(sum, 15, 32767); + + out[out_stride * out_sample++] = sum; + last_sample += int_advance; + samp_frac_num += frac_advance; + if (samp_frac_num >= den_rate) + { + samp_frac_num -= den_rate; + last_sample++; + } + } + + st->last_sample[channel_index] = last_sample; + st->samp_frac_num[channel_index] = samp_frac_num; + return out_sample; +} + + +/* This resampler is used to produce zero output in situations where memory + for the filter could not be allocated. The expected numbers of input and + output samples are still processed so that callers failing to check error + codes are not surprised, possibly getting into infinite loops. */ +static int resampler_basic_zero(SpeexResamplerState *st, spx_uint32_t channel_index, const spx_word16_t *in, spx_uint32_t *in_len, spx_word16_t *out, spx_uint32_t *out_len) +{ + int out_sample = 0; + int last_sample = st->last_sample[channel_index]; + spx_uint32_t samp_frac_num = st->samp_frac_num[channel_index]; + const int out_stride = st->out_stride; + const int int_advance = st->int_advance; + const int frac_advance = st->frac_advance; + const spx_uint32_t den_rate = st->den_rate; + + (void)in; + while (!(last_sample >= (spx_int32_t)*in_len || out_sample >= (spx_int32_t)*out_len)) + { + out[out_stride * out_sample++] = 0; + last_sample += int_advance; + samp_frac_num += frac_advance; + if (samp_frac_num >= den_rate) + { + samp_frac_num -= den_rate; + last_sample++; + } + } + + st->last_sample[channel_index] = last_sample; + st->samp_frac_num[channel_index] = samp_frac_num; + return out_sample; +} + +static int multiply_frac(spx_uint32_t *result, spx_uint32_t value, spx_uint32_t num, spx_uint32_t den) +{ + spx_uint32_t major = value / den; + spx_uint32_t remain = value % den; + /* TODO: Could use 64 bits operation to check for overflow. But only guaranteed in C99+ */ + if (remain > UINT32_MAX / num || major > UINT32_MAX / num + || major * num > UINT32_MAX - remain * num / den) + return RESAMPLER_ERR_OVERFLOW; + *result = remain * num / den + major * num; + return RESAMPLER_ERR_SUCCESS; +} + +static int update_filter(SpeexResamplerState *st) +{ + spx_uint32_t old_length = st->filt_len; + spx_uint32_t old_alloc_size = st->mem_alloc_size; + int use_direct; + spx_uint32_t min_sinc_table_length; + spx_uint32_t min_alloc_size; + + st->int_advance = st->num_rate/st->den_rate; + st->frac_advance = st->num_rate%st->den_rate; + st->oversample = quality_map[st->quality].oversample; + st->filt_len = quality_map[st->quality].base_length; + + if (st->num_rate > st->den_rate) + { + /* down-sampling */ + st->cutoff = quality_map[st->quality].downsample_bandwidth * st->den_rate / st->num_rate; + if (multiply_frac(&st->filt_len,st->filt_len,st->num_rate,st->den_rate) != RESAMPLER_ERR_SUCCESS) + goto fail; + /* Round up to make sure we have a multiple of 8 for SSE */ + st->filt_len = ((st->filt_len-1)&(~0x7))+8; + if (2*st->den_rate < st->num_rate) + st->oversample >>= 1; + if (4*st->den_rate < st->num_rate) + st->oversample >>= 1; + if (8*st->den_rate < st->num_rate) + st->oversample >>= 1; + if (16*st->den_rate < st->num_rate) + st->oversample >>= 1; + if (st->oversample < 1) + st->oversample = 1; + } else { + /* up-sampling */ + st->cutoff = quality_map[st->quality].upsample_bandwidth; + } + + /* Choose the resampling type that requires the least amount of memory */ + use_direct = st->filt_len*st->den_rate <= st->filt_len*st->oversample+8 + && INT_MAX/sizeof(spx_word16_t)/st->den_rate >= st->filt_len; + if (use_direct) + { + min_sinc_table_length = st->filt_len*st->den_rate; + } else { + if ((INT_MAX/sizeof(spx_word16_t)-8)/st->oversample < st->filt_len) + goto fail; + + min_sinc_table_length = st->filt_len*st->oversample+8; + } + if (st->sinc_table_length < min_sinc_table_length) + { + spx_word16_t *sinc_table = (spx_word16_t *)speex_realloc(st->sinc_table,min_sinc_table_length*sizeof(spx_word16_t)); + if (!sinc_table) + goto fail; + + st->sinc_table = sinc_table; + st->sinc_table_length = min_sinc_table_length; + } + if (use_direct) + { + spx_uint32_t i; + for (i=0;iden_rate;i++) + { + spx_int32_t j; + for (j=0;jfilt_len;j++) + { + st->sinc_table[i*st->filt_len+j] = sinc(st->cutoff,((j-(spx_int32_t)st->filt_len/2+1)-((float)i)/st->den_rate), st->filt_len, quality_map[st->quality].window_func); + } + } + st->resampler_ptr = resampler_basic_direct_single; + /*fprintf (stderr, "resampler uses direct sinc table and normalised cutoff %f\n", cutoff);*/ + } else { + spx_int32_t i; + for (i=-4;i<(spx_int32_t)(st->oversample*st->filt_len+4);i++) + st->sinc_table[i+4] = sinc(st->cutoff,(i/(float)st->oversample - st->filt_len/2), st->filt_len, quality_map[st->quality].window_func); + st->resampler_ptr = resampler_basic_interpolate_single; + /*fprintf (stderr, "resampler uses interpolated sinc table and normalised cutoff %f\n", cutoff);*/ + } + + /* Here's the place where we update the filter memory to take into account + the change in filter length. It's probably the messiest part of the code + due to handling of lots of corner cases. */ + + /* Adding buffer_size to filt_len won't overflow here because filt_len + could be multiplied by sizeof(spx_word16_t) above. */ + min_alloc_size = st->filt_len-1 + st->buffer_size; + if (min_alloc_size > st->mem_alloc_size) + { + spx_word16_t *mem; + if (INT_MAX/sizeof(spx_word16_t)/st->nb_channels < min_alloc_size) + goto fail; + else if (!(mem = (spx_word16_t*)speex_realloc(st->mem, st->nb_channels*min_alloc_size * sizeof(*mem)))) + goto fail; + + st->mem = mem; + st->mem_alloc_size = min_alloc_size; + } + if (!st->started) + { + spx_uint32_t i; + for (i=0;inb_channels*st->mem_alloc_size;i++) + st->mem[i] = 0; + /*speex_warning("reinit filter");*/ + } else if (st->filt_len > old_length) + { + spx_uint32_t i; + /* Increase the filter length */ + /*speex_warning("increase filter size");*/ + for (i=st->nb_channels;i--;) + { + spx_uint32_t j; + spx_uint32_t olen = old_length; + /*if (st->magic_samples[i])*/ + { + /* Try and remove the magic samples as if nothing had happened */ + + /* FIXME: This is wrong but for now we need it to avoid going over the array bounds */ + olen = old_length + 2*st->magic_samples[i]; + for (j=old_length-1+st->magic_samples[i];j--;) + st->mem[i*st->mem_alloc_size+j+st->magic_samples[i]] = st->mem[i*old_alloc_size+j]; + for (j=0;jmagic_samples[i];j++) + st->mem[i*st->mem_alloc_size+j] = 0; + st->magic_samples[i] = 0; + } + if (st->filt_len > olen) + { + /* If the new filter length is still bigger than the "augmented" length */ + /* Copy data going backward */ + for (j=0;jmem[i*st->mem_alloc_size+(st->filt_len-2-j)] = st->mem[i*st->mem_alloc_size+(olen-2-j)]; + /* Then put zeros for lack of anything better */ + for (;jfilt_len-1;j++) + st->mem[i*st->mem_alloc_size+(st->filt_len-2-j)] = 0; + /* Adjust last_sample */ + st->last_sample[i] += (st->filt_len - olen)/2; + } else { + /* Put back some of the magic! */ + st->magic_samples[i] = (olen - st->filt_len)/2; + for (j=0;jfilt_len-1+st->magic_samples[i];j++) + st->mem[i*st->mem_alloc_size+j] = st->mem[i*st->mem_alloc_size+j+st->magic_samples[i]]; + } + } + } else if (st->filt_len < old_length) + { + spx_uint32_t i; + /* Reduce filter length, this a bit tricky. We need to store some of the memory as "magic" + samples so they can be used directly as input the next time(s) */ + for (i=0;inb_channels;i++) + { + spx_uint32_t j; + spx_uint32_t old_magic = st->magic_samples[i]; + st->magic_samples[i] = (old_length - st->filt_len)/2; + /* We must copy some of the memory that's no longer used */ + /* Copy data going backward */ + for (j=0;jfilt_len-1+st->magic_samples[i]+old_magic;j++) + st->mem[i*st->mem_alloc_size+j] = st->mem[i*st->mem_alloc_size+j+st->magic_samples[i]]; + st->magic_samples[i] += old_magic; + } + } + return RESAMPLER_ERR_SUCCESS; + +fail: + st->resampler_ptr = resampler_basic_zero; + /* st->mem may still contain consumed input samples for the filter. + Restore filt_len so that filt_len - 1 still points to the position after + the last of these samples. */ + st->filt_len = old_length; + return RESAMPLER_ERR_ALLOC_FAILED; +} + +EXPORT SpeexResamplerState *speex_resampler_init(spx_uint32_t nb_channels, spx_uint32_t in_rate, spx_uint32_t out_rate, int quality, int *err) +{ + return speex_resampler_init_frac(nb_channels, in_rate, out_rate, in_rate, out_rate, quality, err); +} + +EXPORT SpeexResamplerState *speex_resampler_init_frac(spx_uint32_t nb_channels, spx_uint32_t ratio_num, spx_uint32_t ratio_den, spx_uint32_t in_rate, spx_uint32_t out_rate, int quality, int *err) +{ + SpeexResamplerState *st; + int filter_err; + + if (nb_channels == 0 || ratio_num == 0 || ratio_den == 0 || quality > 10 || quality < 0) + { + if (err) + *err = RESAMPLER_ERR_INVALID_ARG; + return NULL; + } + st = (SpeexResamplerState *)speex_alloc(sizeof(SpeexResamplerState)); + if (!st) + { + if (err) + *err = RESAMPLER_ERR_ALLOC_FAILED; + return NULL; + } + st->initialised = 0; + st->started = 0; + st->in_rate = 0; + st->out_rate = 0; + st->num_rate = 0; + st->den_rate = 0; + st->quality = -1; + st->sinc_table_length = 0; + st->mem_alloc_size = 0; + st->filt_len = 0; + st->mem = 0; + st->resampler_ptr = 0; + + st->cutoff = 1.f; + st->nb_channels = nb_channels; + st->in_stride = 1; + st->out_stride = 1; + + st->buffer_size = 160; + + /* Per channel data */ + if (!(st->last_sample = (spx_int32_t*)speex_alloc(nb_channels*sizeof(spx_int32_t)))) + goto fail; + if (!(st->magic_samples = (spx_uint32_t*)speex_alloc(nb_channels*sizeof(spx_uint32_t)))) + goto fail; + if (!(st->samp_frac_num = (spx_uint32_t*)speex_alloc(nb_channels*sizeof(spx_uint32_t)))) + goto fail; + + speex_resampler_set_quality(st, quality); + speex_resampler_set_rate_frac(st, ratio_num, ratio_den, in_rate, out_rate); + + filter_err = update_filter(st); + if (filter_err == RESAMPLER_ERR_SUCCESS) + { + st->initialised = 1; + } else { + speex_resampler_destroy(st); + st = NULL; + } + if (err) + *err = filter_err; + + return st; + +fail: + if (err) + *err = RESAMPLER_ERR_ALLOC_FAILED; + speex_resampler_destroy(st); + return NULL; +} + +EXPORT void speex_resampler_destroy(SpeexResamplerState *st) +{ + speex_free(st->mem); + speex_free(st->sinc_table); + speex_free(st->last_sample); + speex_free(st->magic_samples); + speex_free(st->samp_frac_num); + speex_free(st); +} + +static int speex_resampler_process_native(SpeexResamplerState *st, spx_uint32_t channel_index, spx_uint32_t *in_len, spx_word16_t *out, spx_uint32_t *out_len) +{ + int j=0; + const int N = st->filt_len; + int out_sample = 0; + spx_word16_t *mem = st->mem + channel_index * st->mem_alloc_size; + spx_uint32_t ilen; + + st->started = 1; + + /* Call the right resampler through the function ptr */ + out_sample = st->resampler_ptr(st, channel_index, mem, in_len, out, out_len); + + if (st->last_sample[channel_index] < (spx_int32_t)*in_len) + *in_len = st->last_sample[channel_index]; + *out_len = out_sample; + st->last_sample[channel_index] -= *in_len; + + ilen = *in_len; + + for(j=0;jmagic_samples[channel_index]; + spx_word16_t *mem = st->mem + channel_index * st->mem_alloc_size; + const int N = st->filt_len; + + speex_resampler_process_native(st, channel_index, &tmp_in_len, *out, &out_len); + + st->magic_samples[channel_index] -= tmp_in_len; + + /* If we couldn't process all "magic" input samples, save the rest for next time */ + if (st->magic_samples[channel_index]) + { + spx_uint32_t i; + for (i=0;imagic_samples[channel_index];i++) + mem[N-1+i]=mem[N-1+i+tmp_in_len]; + } + *out += out_len*st->out_stride; + return out_len; +} + +EXPORT int speex_resampler_process_int(SpeexResamplerState *st, spx_uint32_t channel_index, const spx_int16_t *in, spx_uint32_t *in_len, spx_int16_t *out, spx_uint32_t *out_len) +{ + int j; + spx_uint32_t ilen = *in_len; + spx_uint32_t olen = *out_len; + spx_word16_t *x = st->mem + channel_index * st->mem_alloc_size; + const int filt_offs = st->filt_len - 1; + const spx_uint32_t xlen = st->mem_alloc_size - filt_offs; + const int istride = st->in_stride; + + if (st->magic_samples[channel_index]) + olen -= speex_resampler_magic(st, channel_index, &out, olen); + if (! st->magic_samples[channel_index]) { + while (ilen && olen) { + spx_uint32_t ichunk = (ilen > xlen) ? xlen : ilen; + spx_uint32_t ochunk = olen; + + if (in) { + for(j=0;jout_stride; + if (in) + in += ichunk * istride; + } + } + *in_len -= ilen; + *out_len -= olen; + return st->resampler_ptr == resampler_basic_zero ? RESAMPLER_ERR_ALLOC_FAILED : RESAMPLER_ERR_SUCCESS; +} + +EXPORT int speex_resampler_process_float(SpeexResamplerState *st, spx_uint32_t channel_index, const float *in, spx_uint32_t *in_len, float *out, spx_uint32_t *out_len) +{ + int j; + const int istride_save = st->in_stride; + const int ostride_save = st->out_stride; + spx_uint32_t ilen = *in_len; + spx_uint32_t olen = *out_len; + spx_word16_t *x = st->mem + channel_index * st->mem_alloc_size; + const spx_uint32_t xlen = st->mem_alloc_size - (st->filt_len - 1); +#ifdef VAR_ARRAYS + const unsigned int ylen = (olen < FIXED_STACK_ALLOC) ? olen : FIXED_STACK_ALLOC; + spx_word16_t ystack[ylen]; +#else + const unsigned int ylen = FIXED_STACK_ALLOC; + spx_word16_t ystack[FIXED_STACK_ALLOC]; +#endif + + st->out_stride = 1; + + while (ilen && olen) { + spx_word16_t *y = ystack; + spx_uint32_t ichunk = (ilen > xlen) ? xlen : ilen; + spx_uint32_t ochunk = (olen > ylen) ? ylen : olen; + spx_uint32_t omagic = 0; + + if (st->magic_samples[channel_index]) { + omagic = speex_resampler_magic(st, channel_index, &y, ochunk); + ochunk -= omagic; + olen -= omagic; + } + if (! st->magic_samples[channel_index]) { + if (in) { + for(j=0;jfilt_len-1]=WORD2INT(in[j*istride_save]); + } else { + for(j=0;jfilt_len-1]=0; + } + + speex_resampler_process_native(st, channel_index, &ichunk, y, &ochunk); + } else { + ichunk = 0; + ochunk = 0; + } + + for (j=0;jout_stride = ostride_save; + *in_len -= ilen; + *out_len -= olen; + + return st->resampler_ptr == resampler_basic_zero ? RESAMPLER_ERR_ALLOC_FAILED : RESAMPLER_ERR_SUCCESS; +} + +EXPORT int speex_resampler_process_interleaved_float(SpeexResamplerState *st, const float *in, spx_uint32_t *in_len, float *out, spx_uint32_t *out_len) +{ + spx_uint32_t i; + int istride_save, ostride_save; + spx_uint32_t bak_out_len = *out_len; + spx_uint32_t bak_in_len = *in_len; + istride_save = st->in_stride; + ostride_save = st->out_stride; + st->in_stride = st->out_stride = st->nb_channels; + for (i=0;inb_channels;i++) + { + *out_len = bak_out_len; + *in_len = bak_in_len; + if (in != NULL) + speex_resampler_process_float(st, i, in+i, in_len, out+i, out_len); + else + speex_resampler_process_float(st, i, NULL, in_len, out+i, out_len); + } + st->in_stride = istride_save; + st->out_stride = ostride_save; + return st->resampler_ptr == resampler_basic_zero ? RESAMPLER_ERR_ALLOC_FAILED : RESAMPLER_ERR_SUCCESS; +} + +EXPORT int speex_resampler_process_interleaved_int(SpeexResamplerState *st, const spx_int16_t *in, spx_uint32_t *in_len, spx_int16_t *out, spx_uint32_t *out_len) +{ + spx_uint32_t i; + int istride_save, ostride_save; + spx_uint32_t bak_out_len = *out_len; + spx_uint32_t bak_in_len = *in_len; + istride_save = st->in_stride; + ostride_save = st->out_stride; + st->in_stride = st->out_stride = st->nb_channels; + for (i=0;inb_channels;i++) + { + *out_len = bak_out_len; + *in_len = bak_in_len; + if (in != NULL) + speex_resampler_process_int(st, i, in+i, in_len, out+i, out_len); + else + speex_resampler_process_int(st, i, NULL, in_len, out+i, out_len); + } + st->in_stride = istride_save; + st->out_stride = ostride_save; + return st->resampler_ptr == resampler_basic_zero ? RESAMPLER_ERR_ALLOC_FAILED : RESAMPLER_ERR_SUCCESS; +} + +EXPORT int speex_resampler_set_rate(SpeexResamplerState *st, spx_uint32_t in_rate, spx_uint32_t out_rate) +{ + return speex_resampler_set_rate_frac(st, in_rate, out_rate, in_rate, out_rate); +} + +EXPORT void speex_resampler_get_rate(SpeexResamplerState *st, spx_uint32_t *in_rate, spx_uint32_t *out_rate) +{ + *in_rate = st->in_rate; + *out_rate = st->out_rate; +} + +static inline spx_uint32_t compute_gcd(spx_uint32_t a, spx_uint32_t b) +{ + while (b != 0) + { + spx_uint32_t temp = a; + + a = b; + b = temp % b; + } + return a; +} + +EXPORT int speex_resampler_set_rate_frac(SpeexResamplerState *st, spx_uint32_t ratio_num, spx_uint32_t ratio_den, spx_uint32_t in_rate, spx_uint32_t out_rate) +{ + spx_uint32_t fact; + spx_uint32_t old_den; + spx_uint32_t i; + + if (ratio_num == 0 || ratio_den == 0) + return RESAMPLER_ERR_INVALID_ARG; + + if (st->in_rate == in_rate && st->out_rate == out_rate && st->num_rate == ratio_num && st->den_rate == ratio_den) + return RESAMPLER_ERR_SUCCESS; + + old_den = st->den_rate; + st->in_rate = in_rate; + st->out_rate = out_rate; + st->num_rate = ratio_num; + st->den_rate = ratio_den; + + fact = compute_gcd(st->num_rate, st->den_rate); + + st->num_rate /= fact; + st->den_rate /= fact; + + if (old_den > 0) + { + for (i=0;inb_channels;i++) + { + if (multiply_frac(&st->samp_frac_num[i],st->samp_frac_num[i],st->den_rate,old_den) != RESAMPLER_ERR_SUCCESS) + return RESAMPLER_ERR_OVERFLOW; + /* Safety net */ + if (st->samp_frac_num[i] >= st->den_rate) + st->samp_frac_num[i] = st->den_rate-1; + } + } + + if (st->initialised) + return update_filter(st); + return RESAMPLER_ERR_SUCCESS; +} + +EXPORT void speex_resampler_get_ratio(SpeexResamplerState *st, spx_uint32_t *ratio_num, spx_uint32_t *ratio_den) +{ + *ratio_num = st->num_rate; + *ratio_den = st->den_rate; +} + +EXPORT int speex_resampler_set_quality(SpeexResamplerState *st, int quality) +{ + if (quality > 10 || quality < 0) + return RESAMPLER_ERR_INVALID_ARG; + if (st->quality == quality) + return RESAMPLER_ERR_SUCCESS; + st->quality = quality; + if (st->initialised) + return update_filter(st); + return RESAMPLER_ERR_SUCCESS; +} + +EXPORT void speex_resampler_get_quality(SpeexResamplerState *st, int *quality) +{ + *quality = st->quality; +} + +EXPORT void speex_resampler_set_input_stride(SpeexResamplerState *st, spx_uint32_t stride) +{ + st->in_stride = stride; +} + +EXPORT void speex_resampler_get_input_stride(SpeexResamplerState *st, spx_uint32_t *stride) +{ + *stride = st->in_stride; +} + +EXPORT void speex_resampler_set_output_stride(SpeexResamplerState *st, spx_uint32_t stride) +{ + st->out_stride = stride; +} + +EXPORT void speex_resampler_get_output_stride(SpeexResamplerState *st, spx_uint32_t *stride) +{ + *stride = st->out_stride; +} + +EXPORT int speex_resampler_get_input_latency(SpeexResamplerState *st) +{ + return st->filt_len / 2; +} + +EXPORT int speex_resampler_get_output_latency(SpeexResamplerState *st) +{ + return ((st->filt_len / 2) * st->den_rate + (st->num_rate >> 1)) / st->num_rate; +} + +EXPORT int speex_resampler_skip_zeros(SpeexResamplerState *st) +{ + spx_uint32_t i; + for (i=0;inb_channels;i++) + st->last_sample[i] = st->filt_len/2; + return RESAMPLER_ERR_SUCCESS; +} + +EXPORT int speex_resampler_reset_mem(SpeexResamplerState *st) +{ + spx_uint32_t i; + for (i=0;inb_channels;i++) + { + st->last_sample[i] = 0; + st->magic_samples[i] = 0; + st->samp_frac_num[i] = 0; + } + for (i=0;inb_channels*(st->filt_len-1);i++) + st->mem[i] = 0; + return RESAMPLER_ERR_SUCCESS; +} + +EXPORT const char *speex_resampler_strerror(int err) +{ + switch (err) + { + case RESAMPLER_ERR_SUCCESS: + return "Success."; + case RESAMPLER_ERR_ALLOC_FAILED: + return "Memory allocation failed."; + case RESAMPLER_ERR_BAD_STATE: + return "Bad resampler state."; + case RESAMPLER_ERR_INVALID_ARG: + return "Invalid argument."; + case RESAMPLER_ERR_PTR_OVERLAP: + return "Input and output buffers overlap."; + default: + return "Unknown error. Bad error code or strange version mismatch."; + } +} diff --git a/audio/deps/resampler/resample_neon.h b/audio/deps/resampler/resample_neon.h new file mode 100644 index 0000000..d5ca9f9 --- /dev/null +++ b/audio/deps/resampler/resample_neon.h @@ -0,0 +1,188 @@ +/* Copyright (C) 2007-2008 Jean-Marc Valin + * Copyright (C) 2008 Thorvald Natvig + * Copyright (C) 2011 Texas Instruments + * author Jyri Sarha + */ +/** + @file resample_neon.h + @brief Resampler functions (NEON version) +*/ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + - Neither the name of the Xiph.org Foundation nor the names of its + contributors may be used to endorse or promote products derived from + this software without specific prior written permission. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR + CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#if defined(__aarch64__) +static inline int32_t saturate_32bit_to_16bit(int32_t a) { + int32_t ret; + __asm__ ("fmov s0, %w[a]\n" + "sqxtn h0, s0\n" + "sxtl v0.4s, v0.4h\n" + "fmov %w[ret], s0\n" + : [ret] "=r" (ret) + : [a] "r" (a) + : "v0" ); + return ret; +} +#elif defined(__thumb2__) +static inline int32_t saturate_32bit_to_16bit(int32_t a) { + int32_t ret; + __asm__ ("ssat %[ret], #16, %[a]" + : [ret] "=r" (ret) + : [a] "r" (a) + : ); + return ret; +} +#else +static inline int32_t saturate_32bit_to_16bit(int32_t a) { + int32_t ret; + __asm__ ("vmov.s32 d0[0], %[a]\n" + "vqmovn.s32 d0, q0\n" + "vmov.s16 %[ret], d0[0]\n" + : [ret] "=r" (ret) + : [a] "r" (a) + : "q0"); + return ret; +} +#endif +#undef WORD2INT +#define WORD2INT(x) (saturate_32bit_to_16bit(x)) + +#define OVERRIDE_INNER_PRODUCT_SINGLE +/* Only works when len % 4 == 0 and len >= 4 */ +#if defined(__aarch64__) +static inline int32_t inner_product_single(const int16_t *a, const int16_t *b, unsigned int len) +{ + int32_t ret; + uint32_t remainder = len % 16; + len = len - remainder; + + __asm__ volatile (" cmp %w[len], #0\n" + " b.ne 1f\n" + " ld1 {v16.4h}, [%[b]], #8\n" + " ld1 {v20.4h}, [%[a]], #8\n" + " subs %w[remainder], %w[remainder], #4\n" + " smull v0.4s, v16.4h, v20.4h\n" + " b.ne 4f\n" + " b 5f\n" + "1:" + " ld1 {v16.4h, v17.4h, v18.4h, v19.4h}, [%[b]], #32\n" + " ld1 {v20.4h, v21.4h, v22.4h, v23.4h}, [%[a]], #32\n" + " subs %w[len], %w[len], #16\n" + " smull v0.4s, v16.4h, v20.4h\n" + " smlal v0.4s, v17.4h, v21.4h\n" + " smlal v0.4s, v18.4h, v22.4h\n" + " smlal v0.4s, v19.4h, v23.4h\n" + " b.eq 3f\n" + "2:" + " ld1 {v16.4h, v17.4h, v18.4h, v19.4h}, [%[b]], #32\n" + " ld1 {v20.4h, v21.4h, v22.4h, v23.4h}, [%[a]], #32\n" + " subs %w[len], %w[len], #16\n" + " smlal v0.4s, v16.4h, v20.4h\n" + " smlal v0.4s, v17.4h, v21.4h\n" + " smlal v0.4s, v18.4h, v22.4h\n" + " smlal v0.4s, v19.4h, v23.4h\n" + " b.ne 2b\n" + "3:" + " cmp %w[remainder], #0\n" + " b.eq 5f\n" + "4:" + " ld1 {v18.4h}, [%[b]], #8\n" + " ld1 {v22.4h}, [%[a]], #8\n" + " subs %w[remainder], %w[remainder], #4\n" + " smlal v0.4s, v18.4h, v22.4h\n" + " b.ne 4b\n" + "5:" + " saddlv d0, v0.4s\n" + " sqxtn s0, d0\n" + " sqrshrn h0, s0, #15\n" + " sxtl v0.4s, v0.4h\n" + " fmov %w[ret], s0\n" + : [ret] "=r" (ret), [a] "+r" (a), [b] "+r" (b), + [len] "+r" (len), [remainder] "+r" (remainder) + : + : "cc", "v0", + "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23"); + return ret; +} +#else +static inline int32_t inner_product_single(const int16_t *a, const int16_t *b, unsigned int len) +{ + int32_t ret; + uint32_t remainder = len % 16; + len = len - remainder; + + __asm__ volatile (" cmp %[len], #0\n" + " bne 1f\n" + " vld1.16 {d16}, [%[b]]!\n" + " vld1.16 {d20}, [%[a]]!\n" + " subs %[remainder], %[remainder], #4\n" + " vmull.s16 q0, d16, d20\n" + " beq 5f\n" + " b 4f\n" + "1:" + " vld1.16 {d16, d17, d18, d19}, [%[b]]!\n" + " vld1.16 {d20, d21, d22, d23}, [%[a]]!\n" + " subs %[len], %[len], #16\n" + " vmull.s16 q0, d16, d20\n" + " vmlal.s16 q0, d17, d21\n" + " vmlal.s16 q0, d18, d22\n" + " vmlal.s16 q0, d19, d23\n" + " beq 3f\n" + "2:" + " vld1.16 {d16, d17, d18, d19}, [%[b]]!\n" + " vld1.16 {d20, d21, d22, d23}, [%[a]]!\n" + " subs %[len], %[len], #16\n" + " vmlal.s16 q0, d16, d20\n" + " vmlal.s16 q0, d17, d21\n" + " vmlal.s16 q0, d18, d22\n" + " vmlal.s16 q0, d19, d23\n" + " bne 2b\n" + "3:" + " cmp %[remainder], #0\n" + " beq 5f\n" + "4:" + " vld1.16 {d16}, [%[b]]!\n" + " vld1.16 {d20}, [%[a]]!\n" + " subs %[remainder], %[remainder], #4\n" + " vmlal.s16 q0, d16, d20\n" + " bne 4b\n" + "5:" + " vaddl.s32 q0, d0, d1\n" + " vadd.s64 d0, d0, d1\n" + " vqmovn.s64 d0, q0\n" + " vqrshrn.s32 d0, q0, #15\n" + " vmov.s16 %[ret], d0[0]\n" + : [ret] "=r" (ret), [a] "+r" (a), [b] "+r" (b), + [len] "+r" (len), [remainder] "+r" (remainder) + : + : "cc", "q0", + "d16", "d17", "d18", "d19", "d20", "d21", "d22", "d23"); + + return ret; +} +#endif // !defined(__aarch64__) diff --git a/audio/deps/resampler/speex_resampler.h b/audio/deps/resampler/speex_resampler.h new file mode 100644 index 0000000..77f2520 --- /dev/null +++ b/audio/deps/resampler/speex_resampler.h @@ -0,0 +1,216 @@ +/* Copyright (C) 2007 Jean-Marc Valin + + File: speex_resampler.h + Resampling code + + The design goals of this code are: + - Very fast algorithm + - Low memory requirement + - Good *perceptual* quality (and not best SNR) + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions are + met: + + 1. Redistributions of source code must retain the above copyright notice, + this list of conditions and the following disclaimer. + + 2. Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + 3. The name of the author may not be used to endorse or promote products + derived from this software without specific prior written permission. + + THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR + IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES + OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + DISCLAIMED. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, + INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR + SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, + STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN + ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE + POSSIBILITY OF SUCH DAMAGE. +*/ + + +#ifndef SPEEX_RESAMPLER_H +#define SPEEX_RESAMPLER_H +#include "speexdsp_types.h" + +#ifdef __cplusplus +extern "C" { +#endif + +#define SPEEX_RESAMPLER_QUALITY_MAX 10 +#define SPEEX_RESAMPLER_QUALITY_MIN 0 +#define SPEEX_RESAMPLER_QUALITY_DEFAULT 4 +#define SPEEX_RESAMPLER_QUALITY_VOIP 3 +#define SPEEX_RESAMPLER_QUALITY_DESKTOP 5 + +enum { + RESAMPLER_ERR_SUCCESS = 0, + RESAMPLER_ERR_ALLOC_FAILED = 1, + RESAMPLER_ERR_BAD_STATE = 2, + RESAMPLER_ERR_INVALID_ARG = 3, + RESAMPLER_ERR_PTR_OVERLAP = 4, + RESAMPLER_ERR_OVERFLOW = 5, + + RESAMPLER_ERR_MAX_ERROR +}; + +struct SpeexResamplerState_; +typedef struct SpeexResamplerState_ SpeexResamplerState; + +/** Create a new resampler with integer input and output rates. + * @param nb_channels Number of channels to be processed + * @param in_rate Input sampling rate (integer number of Hz). + * @param out_rate Output sampling rate (integer number of Hz). + * @param quality Resampling quality between 0 and 10, where 0 has poor quality + * and 10 has very high quality. + * @return Newly created resampler state + * @retval NULL Error: not enough memory + */ +SpeexResamplerState *speex_resampler_init(spx_uint32_t nb_channels, + spx_uint32_t in_rate, + spx_uint32_t out_rate, + int quality, + int *err); + +/** Destroy a resampler state. + * @param st Resampler state + */ +void speex_resampler_destroy(SpeexResamplerState *st); + +/** Resample an int array. The input and output buffers must *not* overlap. + * @param st Resampler state + * @param channel_index Index of the channel to process for the multi-channel + * base (0 otherwise) + * @param in Input buffer + * @param in_len Number of input samples in the input buffer. Returns the number + * of samples processed + * @param out Output buffer + * @param out_len Size of the output buffer. Returns the number of samples written + */ +int speex_resampler_process_int(SpeexResamplerState *st, + spx_uint32_t channel_index, + const spx_int16_t *in, + spx_uint32_t *in_len, + spx_int16_t *out, + spx_uint32_t *out_len); + +/** Resample an interleaved int array. The input and output buffers must *not* overlap. + * @param st Resampler state + * @param in Input buffer + * @param in_len Number of input samples in the input buffer. Returns the number + * of samples processed. This is all per-channel. + * @param out Output buffer + * @param out_len Size of the output buffer. Returns the number of samples written. + * This is all per-channel. + */ +int speex_resampler_process_interleaved_int(SpeexResamplerState *st, + const spx_int16_t *in, + spx_uint32_t *in_len, + spx_int16_t *out, + spx_uint32_t *out_len); + +/** Set (change) the input/output sampling rates (integer value). + * @param st Resampler state + * @param in_rate Input sampling rate (integer number of Hz). + * @param out_rate Output sampling rate (integer number of Hz). + */ +int speex_resampler_set_rate(SpeexResamplerState *st, + spx_uint32_t in_rate, + spx_uint32_t out_rate); + +/** Get the current input/output sampling rates (integer value). + * @param st Resampler state + * @param in_rate Input sampling rate (integer number of Hz) copied. + * @param out_rate Output sampling rate (integer number of Hz) copied. + */ +void speex_resampler_get_rate(SpeexResamplerState *st, + spx_uint32_t *in_rate, + spx_uint32_t *out_rate); + +/** Set (change) the conversion quality. + * @param st Resampler state + * @param quality Resampling quality between 0 and 10, where 0 has poor + * quality and 10 has very high quality. + */ +int speex_resampler_set_quality(SpeexResamplerState *st, + int quality); + +/** Get the conversion quality. + * @param st Resampler state + * @param quality Resampling quality between 0 and 10, where 0 has poor + * quality and 10 has very high quality. + */ +void speex_resampler_get_quality(SpeexResamplerState *st, + int *quality); + +/** Set (change) the input stride. + * @param st Resampler state + * @param stride Input stride + */ +void speex_resampler_set_input_stride(SpeexResamplerState *st, + spx_uint32_t stride); + +/** Get the input stride. + * @param st Resampler state + * @param stride Input stride copied + */ +void speex_resampler_get_input_stride(SpeexResamplerState *st, + spx_uint32_t *stride); + +/** Set (change) the output stride. + * @param st Resampler state + * @param stride Output stride + */ +void speex_resampler_set_output_stride(SpeexResamplerState *st, + spx_uint32_t stride); + +/** Get the output stride. + * @param st Resampler state copied + * @param stride Output stride + */ +void speex_resampler_get_output_stride(SpeexResamplerState *st, + spx_uint32_t *stride); + +/** Get the latency introduced by the resampler measured in input samples. + * @param st Resampler state + */ +int speex_resampler_get_input_latency(SpeexResamplerState *st); + +/** Get the latency introduced by the resampler measured in output samples. + * @param st Resampler state + */ +int speex_resampler_get_output_latency(SpeexResamplerState *st); + +/** Make sure that the first samples to go out of the resamplers don't have + * leading zeros. This is only useful before starting to use a newly created + * resampler. It is recommended to use that when resampling an audio file, as + * it will generate a file with the same length. For real-time processing, + * it is probably easier not to use this call (so that the output duration + * is the same for the first frame). + * @param st Resampler state + */ +int speex_resampler_skip_zeros(SpeexResamplerState *st); + +/** Reset a resampler so a new (unrelated) stream can be processed. + * @param st Resampler state + */ +int speex_resampler_reset_mem(SpeexResamplerState *st); + +/** Returns the English meaning for an error code + * @param err Error code + * @return English string + */ +const char *speex_resampler_strerror(int err); + +#ifdef __cplusplus +} +#endif + +#endif diff --git a/audio/deps/resampler/speexdsp_types.h b/audio/deps/resampler/speexdsp_types.h new file mode 100644 index 0000000..5f6cc9d --- /dev/null +++ b/audio/deps/resampler/speexdsp_types.h @@ -0,0 +1,28 @@ +/* speexdsp_types.h taken from libogg */ +/******************************************************************** + * * + * THIS FILE IS PART OF THE OggVorbis SOFTWARE CODEC SOURCE CODE. * + * USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS * + * GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE * + * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * + * * + * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2002 * + * by the Xiph.Org Foundation http://www.xiph.org/ * + * * + ********************************************************************/ + +/** + @file speexdsp_types.h + @brief Speex types +*/ +#ifndef __SPEEX_TYPES_H__ +#define __SPEEX_TYPES_H__ + +#include + +typedef int16_t spx_int16_t; +typedef uint16_t spx_uint16_t; +typedef int32_t spx_int32_t; +typedef uint32_t spx_uint32_t; + +#endif